@emilymbender@dair-community.social
Post #1487005
2026-04-20 14:14 UTC
@trochee@dair-community.social ugh. And yes, except:
"A drag-and-drop canvas makes it trivially easy to chain ten LLM calls together and impossibly hard to debug why the eighth one hallucinates on Tuesdays."
Because it "hallucinates" on days that end in y, ofc.
Replies (1)
-
@SecurityWriter@infosec.exchange 2026-04-20 14:27
@emilymbender@dair-community.social @trochee@dair-community.social once again promoting my campaign to stop calling it ‘hallucination’ and start calling it what it is, ‘erroneous output.’ What software error rate is acceptable? 0.1% is a good starting point, 1% a bright red line for a PoC, negligence in the real world. 10%? Don’t be silly. 20%? It’s just working by accident at this point. Real world tests put the average at 45% last month. Correct output just over half the time. A coin flip. The starry-eyed, hopeful, technology-loving childhood me did not want any of this.